RobAgrees's picture
add config.json, readme.md
ee8e7fd verified
|
raw
history blame
489 Bytes

Quantized Dia 1.6B (INT8)

This is a dynamic int8 quantized version of nari-labs/Dia-1.6B. It uses dynamic quantization for lighter deployment and faster inference.

Original model: float16, ~6.4GB
Quantized model: int8 dynamic, ~6.4GB

Uploaded by RobertAgee and [RobAgrees](https://huggingface.co/RobAgrees.

Quantized automatically with PyTorch dynamic quantization in Google Colab.